Stochastic Circuits for Probabilistic Inference
Working with the fabulous Vikash Mansinghka, I've spent a lot of time building novel computing architectures for solving hard problems in probabilistic inference. It turns out that when you embrace, not suppress, stochasticity, you can build systems that are lower power, more robust to noise, and quite composible.
Building fast Bayesian computing machines out of intentionally stochastic, digital parts. Mansinghka and Jonas, Feb 2014, arXiv:1402.4914 [pdf]
Stochastic Architectures for Probabilistic Computation Eric Jonas, MIT BCS PhD Thesis, 2013. [pdf]
Stochastic digital circuits for probabilistic inference. Vikash K Mansinghka, Eric M Jonas, Joshua B Tenenbaum, Massachussets Institute of Technology, Technical Report MITCSAIL-TR-2008-069 [pdf]